Comment 2 for bug 1452619

So I had this idea that maybe the stacking of fragments on the handoffs would be bad, since it'd be common in a multi-node failure scenario and we want to leave the data spread out for durability and availability util the nodes come back up...

... but for the primaries, maybe we don't expect them to have a non-primary index fragment on them (only known way for that to happen organically is for a previously a handoff node becomes a primary in a rebalance) - but even when it does happen it's probably better to get the fragments in their primary locations as quickly as possible to avoid a potentially needless rebuild, or worse a rebuild that will 409.

If you're a primary and someone has a fragment that belongs on you - you eat it - end of story.

Not sure if this diff is the most effective way of communicating that idea, but it accomplishes the idea well enough we could try writing another double_revert_handoff probetest against it and see if we like the behavior.